Systematic comparison of approaches to analyze clustered competing risks data

Area of Science:

• Biostatistics
• Clinical Trials
• Epidemiology

Background:

• Clinical trials often analyze time-to-event data, facing challenges from competing events and multi-center cluster structures.
• Existing analyses frequently address either competing events or clustering, but not both simultaneously.
• There is a need for systematic comparison of methods handling both competing events and data clustering.

Purpose of the Study:

• To systematically compare four statistical approaches for analyzing competing events in the presence of data clustering.
• To evaluate method performance using both a real-world dataset and extensive simulations.
• To identify the most reliable method for applied researchers in complex clinical trial designs.

Main Methods:

• Comparison of four analytical methods: cause-specific Cox model with frailty, Fine and Gray model, and two extensions (Katsahian et al., Zhou et al.).
• Evaluation based on bias, square root of mean squared error (RMSE), and statistical power.
• Utilized a real-life clinical trial dataset and a comprehensive simulation study.

Main Results:

• The Katsahian et al. model exhibited the best performance across nearly all simulated scenarios.
• This model excelled in minimizing bias and RMSE while maximizing statistical power.
• It uniquely provided both unbiased effect estimation and accurate prognosis.

Conclusions:

• The systematic comparison guides researchers in selecting appropriate methods for competing events with clustered data.
• The Katsahian et al. approach is recommended due to its robust performance and dual capability for estimation and prognosis.
• This work enhances the analytical toolkit for complex clinical trial data analysis.